🚀Cuff-KT: Tackling Learners' Real-time Learning Pattern Adjustment via Tuning-Free Knowledge State-Guided Model Updating (KDD2025)

PyTorch implementation of Cuff-KT.

🌟Data and Data Preprocessing

Place the assist15, assist17, comp, xes3g5m, and dbe-kt22 source files in the dataset directory, and process the data using the following commands respectively:

python preprocess_data.py --data_name assistments15
python preprocess_data.py --data_name assistments17
python preprocess_data.py --data_name comp
python preprocess_data.py --data_name xes3g5m
python preprocess_data.py --data_name dbe_kt22

You can also download the dataset from dataset and place it in the dataset directory.

The statistics of the five datasets after processing are as follows:

Datasets	#learners	#questions	#concepts	#interactions
assist15	17,115	100	100	676,288
assist17	1,708	3,162	411	934,638
comp	5,000	7,460	445	668,927
xes3g5m	5,000	7,242	1,221	1,771,657
dbe-kt22	1,186	212	127	306,904

➡️Quick Start

Installation

Git clone this repository and create conda environment:

conda create -n cuff python=3.11.9
conda activate cuff
pip install -r requirements.txt 

Alternatively, download the environment package from environment and execute the following commands in sequence:

• Navigate to the conda installation directory: /anaconda (or miniconda)/envs/
• Create a folder named cuff in that directory
• Extract the downloaded environment package to the conda environment using the command:

tar -xzvf cuff.tar.gz -C /anaconda (or miniconda)/envs/cuff/
conda activate cuff

Training & Testing

You can execute experiments directly using the following commands:

• Controllable Parameter Generation

CUDA_VISIBLE_DEVICES=0 python main.py --exp intra --model_name [dkt, atdkt] --data_name [assistments15, assistments17, comp, xes3g5m, dbe_kt22] --method cuff --rank 1 --control [ecod, pca, iforest, lof, cuff] --ratio [0, 0.2, 0.4, 0.6, 0.8, 1]
CUDA_VISIBLE_DEVICES=0 python main.py --exp intra --model_name [dkvmn, stablekt, dimkt, diskt] --data_name [assistments15, assistments17, comp, xes3g5m, dbe_kt22] --method cuff --rank 1 --control [ecod, pca, iforest, lof, cuff] --ratio [0, 0.2, 0.4, 0.6, 0.8, 1] --convert True

• Tuning-Free and Fast Prediction
• • baselines

CUDA_VISIBLE_DEVICES=0 python main.py --exp [intra, inter] --model_name [dkt, atdkt] --data_name [assistments15, assistments17, comp, xes3g5m, dbe_kt22]
CUDA_VISIBLE_DEVICES=0 python main.py --exp [intra, inter] --model_name [dkvmn, stablekt, dimkt, diskt] --data_name [assistments15, assistments17, comp, xes3g5m, dbe_kt22] --convert True
CUDA_VISIBLE_DEVICES=0 python main.py --exp [intra, inter] --model_name [dkt, atdkt] --data_name [assistments15, assistments17, comp, xes3g5m, dbe_kt22] --method [fft, adapter, bitfit]
CUDA_VISIBLE_DEVICES=0 python main.py --exp [intra, inter] --model_name [dkvmn, stablekt, dimkt, diskt] --data_name [assistments15, assistments17, comp, xes3g5m, dbe_kt22] --method [fft, adapter, bitfit]  --convert True

• • cuff-kt

CUDA_VISIBLE_DEVICES=0 python main.py --exp [intra, inter] --model_name [dkt, atdkt] --data_name [assistments15, assistments17, comp, xes3g5m, dbe_kt22] --method cuff --rank 1
CUDA_VISIBLE_DEVICES=0 python main.py --exp [intra, inter] --model_name [dkvmn, stablekt, dimkt, diskt] --data_name [assistments15, assistments17, comp, xes3g5m, dbe_kt22] --method cuff --rank 1 --convert True

• Flexible Application

CUDA_VISIBLE_DEVICES=0 python main.py --exp [intra, inter] --model_name [dkt, atdkt] --data_name [assistments15, assistments17, comp, xes3g5m, dbe_kt22] --method cuff+ --rank 1
CUDA_VISIBLE_DEVICES=0 python main.py --exp [intra, inter] --model_name [dkvmn, stablekt, dimkt, diskt] --data_name [assistments15, assistments17, comp, xes3g5m, dbe_kt22] --method cuff+ --rank 1 --convert True

⚠️Citation

If you find our work valuable, we would appreciate your citation:

@misc{zhou2025cuffkttacklinglearnersrealtime,
      title={Cuff-KT: Tackling Learners' Real-time Learning Pattern Adjustment via Tuning-Free Knowledge State Guided Model Updating}, 
      author={Yiyun Zhou and Zheqi Lv and Shengyu Zhang and Jingyuan Chen},
      year={2025},
      eprint={2505.19543},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2505.19543}, 
}